Michał Lipiński

It was always more than about performance. I started Solid because I liked the mental model of Knockout more than React. Retained mode fit with the DOM better than this immediate mode abstration that React put out there. I needed to focus on performance initially to undo the damage caused by DOM slow, VDOM fast. A lot of benchmarks looked at diffing speeds when diffing wasn't what users primarily need to be doing. This starts pre-React Hooks. React adopted this language later and honestly for some people on this side of things that was enough. But it became more obvious just how different these these things end up being. The key difference is React conflates state with UI organization. On change of state the declared body (and descendants) need to re-run if things change. With Signals it is only the computations that need to re-run. It looks similar but it means that state runs through components rather than are the components. I do think there is some preference here. I'm not saying everyone will find Signals better fit for their mental model. But we shouldn't be just suggesting it's a performance thing. People will point to this run-once mentality being simpler to understand once there are multiple isolated states involved, but the real power is that you can describe more complicated states in a hoisted manner and have isolated updates. If something changes in a complex data structure in React it has no choice but to re-run all the way down that UI tree. Memoizationn can prune branches but it needs to run all the way from declaration to usage. Yes Signals need to run along their portion of the graph but that is scaled on the derivation of the data not the UI. More so if your application is data dependency aware it opens up different sorts of problem space solutions. Granular partial Hydration/Resumability. I was hired at eBay because they recognized the importance of being able to separate the state from the UI which could stay on the server. My recent research has been into self healing graphs and granular derivation. These are areas where we will see tangible differences when it comes scaling of larger granular/incremental data systems. Things like sync engines. I'm glad that @ryanflorence is asking these questions because these are common thoughts, and it provides an opportunity to refine education in these areas, because unlike React the communities here smaller and have had less time to really refine the message.